The Level Past L4.

Anthropic published their deployment guide for Claude Cowork at the end of April. It’s titled Deploying Claude across your organization, and it does what a good piece of enterprise documentation should do — it gives the reader a map. There’s a maturity model with five levels. A six-month rollout framework. Case studies from Thomson Reuters, Zapier, and Jamf. It’s the most useful single piece of writing I’ve seen come out of an AI lab on the question of how do we actually use this thing at work.

It’s also, very obviously, written for enterprises.

That’s not a complaint. Enterprises are who buy Enterprise plans, and Anthropic is doing exactly what a serious vendor should do — meeting their largest customers where they live. But the map they drew has an edge, and the edge is where I’ve spent the last nine months working.

This is a short note about what happens past the edge.

The published ladder

Anthropic’s model has five levels, L0 through L4.

L0 is using Claude like a chatbot — ask a question, get an answer.
L1 is having Claude produce something real against your files: a deck, a memo, a spreadsheet.
L2 is taking one repeatable workflow and turning it into a skill — a markdown file Claude runs the same way every time.
L3 is bundling skills, scheduling them, and letting agents run on a cadence.
L4 is the destination they describe: every department in your org runs a curated plugin bundle, admin-provisioned, with skills + connectors + scheduled agents all working together.

It’s a clean ladder. Each rung builds on the last. The implicit promise is that if you climb it, you get to L4 and you’re done.

If you’re a Fortune 500 with departments and a security team and a champion to lead the rollout, that’s true. The ladder ends where their reference customer’s structure ends.

But there are about 30 million small and owner-operated businesses in the United States that don’t have departments. They have functions, and one human running all of them. For those businesses, the ladder doesn’t end at L4. It compresses past it.

What compression looks like

I’ll be specific.

I work alongside Jenn Dieas, the founder of Pretty Creative Co. and the operating partner at a multi-location airbrush spray tan business called Glowout. Nine months ago, Glowout was at L0. Jenn used ChatGPT occasionally. The business ran out of her head — the training, the brand voice, the retail script, the financial intuition, the partnership outreach. All of it lived in one person.

We started climbing.

L1 looked like a weekly booking report pulled from Square and a monthly P&L narrative written from her QuickBooks export. Real deliverables, real systems, no new tools for the staff to learn.

L2 looked like encoding her training SOP and her retail-attach script — a one-question consultative pattern she’d been using for years that her technicians weren’t executing consistently — as skills Claude could run identically every time.

L3 was the scheduled stuff: a Monday-morning operator brief at 7:30 AM, a Friday retail-attach report by technician, continuous DM and review monitoring with drafted responses queued for approval.

L4 was where the skills and agents got bundled into proper functional plugins — operations, marketing, finance, customer service. By the end of month nine, every major function in the business had a dedicated plugin bundle, exactly as Anthropic’s map describes.

And then we did something the map doesn’t describe.

This is what I’ve started calling L5 — the Compressed Operator Stack. It’s what L4 becomes when the business doesn’t have departments to distribute across. Instead of distributing plugins across an org, you stack all the plugins into one operator. The owner-operator endgame is one human, one machine, every system live, all functions in one session.

Why the map stops at L4

The reason Anthropic’s published ladder ends at L4 isn’t an oversight. It’s a structural artifact of who they wrote it for. When your reference customer has departments, the natural terminus is each department running its own bundle. There’s no obvious reason to compress past that — Legal doesn’t need Finance’s plugin running in its session.

But when your reference customer is an owner-operator, the math inverts. Distributing across non-existent departments isn’t useful. Compressing into one operator is the whole point. The owner is the integration layer the business used to need a half-dozen junior hires to be.

This isn’t a theoretical claim. Glowout has been operating this way for nine months. Retail attach rate moved decisively above industry benchmark. Training drift across locations resolved because the SOP is the same skill regardless of who runs the session. Jenn’s calendar reclaimed the hours she used to spend being the manual integration layer. Tribal knowledge fully encoded — she can step away for a week and the business keeps running.

That’s not L4. That’s the next thing.

What this means

Three implications, in the order they matter:

For owner-operated businesses: the destination isn’t the same as the destination Anthropic drew. You’re not trying to roll Claude across a department. You’re trying to compress your entire operating layer into one daily session. The work it takes to get there is the same work — climb the ladder — but the endpoint is different and you should plan for it.

For the people deploying Claude into small businesses: L5 is the standard you should be working toward, not L4. If you stop at L4 you’ve built a department’s worth of plugins for a business that doesn’t have departments. The compression step is the one that actually changes the operator’s daily life.

For the framework itself: I expect Anthropic will eventually formalize something like this, because the small-business segment is enormous and the pattern is consistent. Whatever they name it, this is what it is: the level where all functional plugins stack into one operator session and the human becomes the apex orchestrator instead of the integration layer.

A closing note

I don’t think there’s anything proprietary about the L5 pattern. It’s the natural shape that emerges when you take Anthropic’s published model and apply it to a business that doesn’t fit the published assumptions. Anyone working with owner-operators long enough is going to converge on it.

What I do think is that nobody has named it yet, which is why I’m naming it.

If you’re an owner-operator looking at the Anthropic playbook and wondering what it means for a business shaped like yours, the short answer is: keep climbing past where their map ends. The compressed operator stack is the level where the work compounds, and it’s a level you can reach in nine months if someone runs the deployment correctly.

That’s the work. We’ve been doing it. Now it has a name.